INFLUENCE OF CERTAIN ELEMENTS ON STEEL. 389
error exceeding 3 000 pounds. In the basic steel there are 26 groups with over 4 heats, and 25 are within 1500 Ib., and 17 within 1000 pounds. There is 1 group of 53 heats, averaging about 0.1 per cent, of carbon, which shows an error of + 1810 pounds. Putting aside mathematical errors which can hardly be present in this investigation (owing to repeated checking of the totals at each separate rearrangement), it may appear probable that this group contains some abnormal bars, and it may also appear possible that some of the other large groups show a,n agreement through the averaging of bars showing wide differences among themselves.
Table XVII-Z gives some information on this point. Every group in Table XVII-Y comprising more than 50 heats and containing less than 0.225 per cent, of carbon is subdivided so as to have only one-half the former variation in manganese. Thus, if a group comprised heats ranging from 0.4 to 0.49 per cent, of manganese, it is subdivided into one group ranging from 0.4 to 0.44 per cent., and another from 0.45 to 0.49 per cent. If the original group were an average of unlike units, it is probable that the fact would be made manifest by a wide difference between the two parts, but in no case is such a difference discernible.
In the case of the one group composed of 53 heats before mentioned, a more extended analysis is given in Table XVII-Z. It has been divided into 10 parts, the first containing only those heats that contained 0.4 per cent, of manganese, the second those with 0.41 per cent, of manganese, and so on. The number of heats in some of the subdivisions is small, and complete regularity could hardly be expected, but in these 10 subdivisions the smallest difference between the strength as calculated by the formula and the strength as found by the testing-machine is + 723 Ib., and the greatest is -)-2729 Ib., so that the deviation of this group from the general rule is not due to one or two abnormal bars. With this one exception, the cause of which remains unexplained, all the large groups show a difference of less than 1500 Ib. between the actual and the calculated strength, which is perhaps as close an agreement as could be expected.
A careful analysis was made to discover whether anything could be learned from the so-called errors. If, for instance, the groups of low carbon had shown a considerable and Uniform minus error and the groups of high carbon had uniformly shown a similar plus